These are chat archives for dropbox/pyston

16th
Dec 2016
For example "fasta"
Marius Wachtler
@undingen
Dec 16 2016 15:19
yes looks like we only run this benchmarks http://speed.pyston.org/changes/, I guess kmod removed them because the perf was varying alot
but I recently run them manually on a slightly modified master branch (will merge this changes soon):
interp2.py                      2.32s (18.2MB) cpython: 3.99s (-41.7%)
raytrace.py                     4.55s (25.0MB) cpython: 8.31s (-45.2%)
nbody.py                        7.33s (18.4MB) cpython: 12.24s (-40.2%)
fannkuch.py                     5.69s (18.0MB) cpython: 10.99s (-48.2%)
chaos.py                       10.59s (34.1MB) cpython: 21.52s (-50.8%)
fasta.py                        2.90s (18.3MB) cpython: 3.88s (-25.4%)
pidigits.py                     2.41s (18.0MB) cpython: 6.27s (-61.5%)
richards.py                     0.50s (20.6MB) cpython: 1.43s (-64.8%)
deltablue.py                    0.78s (22.4MB) cpython: 1.49s (-47.7%)
lesshaste
@lesshaste
Dec 16 2016 15:22
that would be great
looks a little sad currently to have all those tick boxes that give blank outputs
Marius Wachtler
@undingen
Dec 16 2016 15:23
I agree
:-)
lesshaste
@lesshaste
Dec 16 2016 15:24
while I have you.. I have slightly lost track of pyston's aims . The blogs don't discuss making general algorithmic code fast but from the figures you just gave it looks like you are managing that
I mean the pyston blog
I suppose once all the benchmarks are up it will be a lot clearer what is fast and what isn't
Marius Wachtler
@undingen
Dec 16 2016 15:37
we tend to ignore this benchmarks because they don't seem to be a good indicator for the web server workload. This benchmarks spend most of the time in doing the same thing e.g. list slicing so small changes to this specific parts change the benchmark numbers a lot. But I see probably the next blog post should include this benchmark numbers so that people can see that we are doing quite well on them too :-).
lesshaste
@lesshaste
Dec 16 2016 15:42
right.. I for example have no interest at all in web serving :)
(well I do read web pages...)
Marius Wachtler
@undingen
Dec 16 2016 15:43
:-), I will make sure to include them from now on
lesshaste
@lesshaste
Dec 16 2016 15:43
great! There is a whole world of python coders out there (like me) who are arguably using the wrong language
that is we don't really use the dynamic features of the language and write mostly algorithmic code and would like it to be a little faster. There is numba, but you have to invest in the anaconda infrastructure, and pypy, which is great but dies as soon as you want to use anything that isn't pure python
I assume you know all this of course :)
why don't we code in a more suitable language I hear you ask? Because we can code our ideas in an hour in python without having to worry about any low level details at all. Basically we can worry about our ideas instead of worrying about the code.
(I'll stop now :) )
Marius Wachtler
@undingen
Dec 16 2016 15:52
:-). if you have time I definitely suggest to try pyston out on your workload. We have a lot smaller and bigger and changes over cpython (beside the JIT :-P, e.g. using gmp for the "long" type) which could help in your code too.
lesshaste
@lesshaste
Dec 16 2016 16:01
I will do.. should I report what I find somewhere? Here?
what is the status of working with numpy etc and pyston?
in particular, do I need to do anything to get it to work?
Sun
@Daetalus
Dec 16 2016 16:05
In theory, there has no issue with NumPy, SciPy, Scikit-learn test suite with Pyston. (If have, please let me know or create an issue on Pyston github repo, thanks). All you need to do is use Pyston customized Cython(if your Cython older than 0.25) or the newest Cython 0.25
lesshaste
@lesshaste
Dec 16 2016 16:05
ok thanks.. I saw the pyston patches in cython
I'll report back here when I get a chance to play with it
Sun
@Daetalus
Dec 16 2016 16:06
I mean Pyston could work with Latest version of NumPy, SciPy, Scikit-learn. but there has few issues which not related with Pyston. But related with its own.
their own.
lesshaste
@lesshaste
Dec 16 2016 16:08
thanks
Sun
@Daetalus
Dec 16 2016 16:17
np